A crucial aspect of recommending news articles is the relevance of currentness of articles. Every day, news portals add plenty of new articles. Typically, users are more interested in recently published articles (or articles that provide background information to recently published articles) than in older ones. That leads to the demand to continuously adapt the set of recommendable items in a recommender system. In this paper, we share our experiences with the usage of the generic and open source Data Stream Management System (DSMS) Odysseus as a Recommender System in the CLEF NewsREEL 2017. We continuously calculate the currently most read articles based on a stream of impression events. Our approach uses operators of a stream-based variant of the relational algebra that respects validity intervals of events. This allows us to continuously calculate the K most popular articles (Top-K set regarding to the number of views) in a sliding time window based on well established relational operations. The flexible composition of operators allows us to variate, e.g., the grouping of impressions to get different recommendation sets for different user groups or the exclusion of articles the user already knows.
[1]
Martha Larson,et al.
Benchmarking News Recommendations: The CLEF NewsREEL Use Case
,
2016,
SIGF.
[2]
Jennifer Widom,et al.
The CQL continuous query language: semantic foundations and query execution
,
2006,
The VLDB Journal.
[3]
Jennifer Widom,et al.
Models and issues in data stream systems
,
2002,
PODS.
[4]
JÜRGEN KRÄMER,et al.
Semantics and implementation of continuous sliding window queries over data streams
,
2009,
TODS.
[5]
Frank Hopfgartner,et al.
CLEF 2017 NewsREEL Overview: A Stream-Based Recommender Task for Evaluation and Education
,
2017,
CLEF.
[6]
Timo Michelsen,et al.
Odysseus: a highly customizable framework for creating efficient event stream management systems
,
2012,
DEBS.